Human-like Decision-Making System for Overtaking Stationary Vehicles Based on Traffic Scene Interpretation

Area of Science:

• Artificial Intelligence
• Robotics
• Computer Vision

Background:

• Autonomous vehicles (AVs) face challenges navigating urban environments with stationary vehicles.
• Existing methods for overtaking decisions have limitations, including undesired maneuvers or deadlock situations.

Purpose of the Study:

• To develop a Deep Neural Network (DNN) model for human-like overtaking maneuver decisions in autonomous vehicles.
• To overcome limitations of current approaches by analyzing significant decision factors.

Main Methods:

• Extracted significant traffic-related and intention-related decision factors from urban traffic scenes.
• Designed and implemented a DNN model utilizing these factors as inputs for decision-making.
• Trained the DNN model to generate overtaking maneuver decisions mimicking human drivers.

Main Results:

• The extracted decision factors significantly improved the learning performance of the DNN model.
• The proposed system enabled autonomous vehicles to generate more human-like overtaking maneuver decisions.
• Validation confirmed the model's effectiveness in various urban traffic scenarios.

Conclusions:

• The DNN model, incorporating extracted decision factors, provides a robust solution for AV overtaking maneuvers.
• This approach enhances the safety and efficiency of autonomous navigation in complex urban settings.
• The study contributes to more sophisticated and human-like decision-making capabilities for autonomous vehicles.